Is Chess "Thinking?"

Today in Philosophy class the subject was Newell and Simon's 1975 Turing Award Lecture - the one outlining the Physical Symbol System Hypothesis. This hypothesis, for those who don't know, is the idea that a necessary and sufficient condition for an intelligent system (such as the human brain, or some hypothetical thinking robot) is that it be a physical symbol system - which means a system that combines symbols (meant here as physical placeholders for concepts in a kind of calculator) into expressions and performs operations on these expressions which yield other expressions. A good example of such a system would be a pocket calculator. A calculator simply carries out physical instructions within a symbol system that happens to correspond to real math. For Newell and Simon, this is the meaning of intelligence (though they might not have accepted that a calculator is "intelligent," really - they are careful to say a system "of a certain size" - meaning of a certain complexity).

Of course this theory isn't without controversy, but something about it seems likely to be true. Logical and mathematical reasoning, after all, can very clearly be modeled by Physical Symbol Systems.

One example that was brought up in class was that of Deep Blue beating Gary Kasparov at Chess. This seemed like a clear example of a machine - a physical symbol system - exhibiting intelligent behavior, albeit in a limited domain. However, one of the students said it was a "bad example," without explanation.

So I asked him what was wrong with the example. His reply was interesting. His objection was that when brainscans are done of master players playing Chess, they don't pattern the same as brain scans done on people doing complex mathematical reasoning.

I asked for clarification on what he meant. Did this mean that Deep Blue didn't demonstrate "intelligence" in the relevant definition because it wasn't solving the problem in the same way as a human? Or did it mean that Chess itself is not intelligent activity.

Astonishingly, perhaps, the answer was the latter! It was asserted that Chess itself isn't an "intelligent" activity. Rather, according to these brain scan studies, it's actually just a highly intuitive form of spacial reasoning - as opposed to a search through some kind of data space for the legal move that seems likely to lead to the best results. Not a rational activity, but a guessing game of some kind?

Some grand masters would agree with this. I remember reading once in a book about Shogi(a Japanese Chess-like game) that at some level when Shogi masters make a move it isn't because they reasoned it out, necessarily - it's because "it's the move they like." Chess masters seem to be doing something similar.

On reflection, though, this seems like a stupid way of looking at it, for two reasons.

First, Chess has to be an intelligent activity. Any operational definition of "intelligent activity" we could come up with would flag Chess. It involves responses to an environment, one can increase one's skill through a process of feedback, etc. All the trappings are there. That some players seem not use use their "intelligence" to play it seems irrelevant.

Second, even if Chess isn't an "intelligent" activity for humans, if a computer is able to play it well (and Deep Blue did, after all, beat history's best player) using methods of reasoning, then surely the computer has demonstrated "intelligent" behavior?

It occured to me, in other words, that the distinction I was drawing between "not playing like a human" and "Chess isn't intelligent" was actually an illusion. These two responses amount to the same: that using Deep Blue is a bad example because it doesn't play Chess in the same (apparently purely intuitive) way that a human does.

(Although, to be fair, it does rather all depend on what you mean by "thinking." In our society there is a sharp divide between intellectual work and physical work, and clearly when we think of Chess we are thinking of intellectual work. There may, however, be some reason to doubt this division. Competing in martial arts, after all, requires a high degree of skill. Slowed down (or translated into video game form), martial arts is actually largely a strategy game. Of course, in real time it doesn't feel like one because everything is done on a purely reactionary level - like instinct, but maybe not quite - players do speak of acting deliberately. For the average Chess player, Chess is probaby played exactly the way the computer played it - by hypothesizing what will happen when certain moves are made. Grand masters maybe do it a bit more like martial arts experts fight - by simply reacting with what "feels right." Obviously there is some difference, though, as all the external behavior exhibited by professional Chess players would suggest that they are, in fact, in deep intellectual-style thought. They rub their heads, sigh, bunch up their brows, and, most importantly, take a long time making their decisions. But maybe on some level there is a real similarity between the kind of intelligence that goes on in a fight and the kind of intelligence that goes on in a Chess game. I myself do not see how the brain can be completely divorced from the decisions a fighter makes in the ring.)

In any case, I don't buy it. Deep Blue very definitely is a demonstration that computers possess intelligence. It doesn't yet seem to function on the same level as human intelligence. There is no self-awareness, and there is no intuition/emotion. Speaking of equivalent intelligence at this point is premature, to say the least.

However, I don't think the Deep Blue example should simply be shrugged off. The computer displayed highly sophisticated cognitive ability in winning that game. If nothing else, it's a prima facie reason to expect that someday someone will design a truly "intelligent" machine.

In other good news, making this post led to the discovery of GNU Shogi, a freeware Shogi program for Unix/Linux that I will attempt to install this weekend!